Cat log remote tidy fix.

[hjoliver]

Follow-up aborted #2672 and earlier attempts at avoiding orphaned tail processes on job hosts.

No need for watch-and kill threads or killing process groups. Just kill invoked subprocesses (either ssh command reinvocation or tail-follow command) if main process's PPID or PPPID (etc) changes.

Example: on host A: cylc cat-log --host=B <suite> <task-on-C>

• => on host A: cat-log spawns subprocess ssh B "cylc cat-log <suite> <task-on-C>"
• => on host B: cat-log spawns subprocess ssh C "cylc cat-log --remote <suite> <task-on-C>"
• => on host C: cat-log spawns subprocess tail -f <task-on-C>.out

Ctrl-C (or exit GUI viewer) on host-A:

• ssh from A to B dies
• => on B: cat-log detects the previous, and kills its ssh subprocess to C
• => on C: cat-log detects the previous, and kills its tail subprocess, then exits normally as finished

[hjoliver]

I haven't seen any clean-up failures with this implementation, after much testing - with bash-4 (remote exec) and tcsh (no remote exec) as login shell.

[hjoliver]

(If this doesn't do it, much knashing of teeth will ensue 😬 )

[hjoliver]

Note edit mode doesn't really work from a third host (same on master, however). It results in (e.g.): ssh vim <file>. [Update: I disabled this].

[matthewrmshin]

Looks good. Tested as working in my environment for a combination of scenarios.

[matthewrmshin]

@oliver-sanders please sanity check.

[hjoliver]

Phew! Release tomorrow...

[oliver-sanders]

Sorry! Logic seems sane, one issue with the -m change.

[oliver-sanders]

There are later usages of options.mode which means cylc cat-log <suite> -m tail results in traceback.

[hjoliver]

@oliver-sanders - good spotting, thanks. Will make the long-time-coming release on approval.

[oliver-sanders]

Looks OK, tested locally, lets hope we've got it this time!

• Travis-CI failure was spurious, test passes locally.
• Codacy would prefer /bin/ps to ps, this is something we can address at a later date if of importance.